Research On Key Technologies Of Battlefield Wireless Network Situation Awareness

The battlefield wireless network(BWN)used to support information transmission has unprecedentedly expanded the combat scope and significantly improved the operational efficiency of the modern battlefield,which has promoted the continuous development of the new military reforms centered on informatization.Realizing the effective situation awareness of the battlefield wireless network,which is considered a critical capability in mastering the initiative of information warfare,can provide auxiliary decisionmaking support for commanders.The battlefield data present the characteristics of massive,complex,and natural distortion,which brings great challenges to the battlefield wireless network situation awareness(BWNSA).Based on the theory of matrix completion and tensor completion,this thesis focuses on the BWNSA method from four perspectives,including node localization,topology inference,traffic estimation,and traffic prediction,on the premise that it does not participate in target network communication and does not demodulate the signal,which provides some new technical means for the BWNSA under the condition of complex noise and incomplete information.(1)Aiming at the BWN node localization problem under the condition of complex noise and missing data,a node localization algorithm based on extended Bregman divergence is proposed by utilizing matrix completion(MC)technology.Based on the general mathematical description of the BWN node localization problem,which is divided into two stages: Euclidean distance matrix(EDM)recovery and node coordinate mapping.In the EDM recovery stage,leveraging the natural low-rank of EDM,the EDM recovery problem is formulated as the MC problem under the condition of complex noise,the MC model is established based on the multi-regularization technique.The vector space linear Bregman iterative algorithm is extended to matrix space to solve the resulting optimization problem.In the node coordinate mapping stage,based on the recovered EDM,the coordinate mapping of unknown nodes is achieved by using the multi-dimensional scaling method and beacon node coordinates.The experimental results show the superior performance and excellent noise tolerance of the proposed method.(2)Aiming at the problem that it is difficult for non-collaborators to obtain network topology by using traditional topology inference(TI)methods in BWN,according to various data modeling methods,a topology inference technology for BWN based on binary matrix completion/tensor completion is proposed.On the one hand,the TI problem at t time is abstracted as a binary MC problem.The concept of positive and unlabeled(PU)learning is introduced,therefore,a PU matrix completion model is proposed based on the non-convex regularization technique,with the inference error of the model being derived theoretically.Two acceleration strategies are utilized to improve the proximal algorithm to solve the resultant non-convex optimization model.On the other hand,the TI problem in a certain period [0,T] is formulated as a binary tensor completion(TC)problem.The concept of PU learning is introduced into TC for the first time,and a general framework for PU tensor completion is proposed.The intrinsic optimality of the framework is proved from two perspectives of the upper bound and minimax lower bound of inference error,and the resulting optimization problem is solved based on alternating direction method of multipliers(ADMM)method.A series of experiments on synthetic data sets and real data sets verify the superior performance of the two proposed BWN topology inference methods.(3)Aiming at the problem that it is difficult to obtain the traffic information between all nodes by direct measurement in BWN,the traffic estimation technology based on TC and the traffic real-time estimation technology based on dynamic MC/streaming TC are proposed.On the one hand,according to the natural representation of the traffic data,the static traffic data in the continuous observation period is constructed as a ”traffic tensor”,and the traffic estimation problem is formulated as a TC problem.On the basis of defining the new tensor p-shrinkage nuclear norm,the corresponding TC model is established,with the estimation error being derived theoretically.An ADMM algorithm based on adaptive momentum scheme acceleration is proposed to solve the resulting nonconvex optimization problem.On the other hand,for the real-time estimation of BWN traffic,the traffic data are constructed in the form of dynamic matrix and streaming tensor respectively.For the former,the real-time traffic estimation problem is abstracted as a dynamic MC problem,and the noise is fitted progressively by a specific Gaussian mixture model(GMM)at each time slot.On the basis of defining the GMM regularization operator,the corresponding dynamic MC model is established and solved based on the expectation-maximization algorithm.For the latter,the real-time traffic estimation problem is formulated as a streaming TC problem,with a streaming TC framework based on dynamic CANDECOMP/PARAFAC(CP)decomposition being established,which is solved based on ADMM algorithm.(4)Aiming at the problem that it is difficult to achieve accurate prediction for BWN traffic by traditional methods due to its burst and irregularity,according to the data integrity,a BWN traffic prediction method based on tensor robust principal component analysis(TRPCA)is proposed respectively.Based on TRPCA technology,the method decomposes traffic data into periodic component and burst component,in which periodic component describes the long-range dependence of BWN traffic,and burst components depicts the burst and irregularity of BWN traffic.The two components are predicted by the Prophet model and Gaussian process regression model,respectively.The experimental results demonstrate the superior performance and excellent stability of the proposed prediction methods.In conclusion,the results of this thesis not only enrich the MC/TC theory,but also expand the application of MC and TC in BWNSA,and provide a novel idea for the research in the field of situation awareness.